The gpt-oss models leverage our state-of-the-art approaches to safety training. We used deliberative alignment and an instruction hierarchy during post-training to help the models refuse unsafe prompts and defend against prompt injections, as well as pre-training interventions.

→ View original post on X — @openai